CCAR-F scenarios › Multi-Agent Research System · Question 30 of 60

Scenario 3 · Multi-Agent Research System

You're building a research system on the Claude Agent SDK in which a coordinator agent delegates work to specialized subagents: one searches the web, one analyzes documents, one synthesizes findings, and one generates reports. Given a topic, the system produces a thorough report with citations.

Context Management & Reliability · Select TWO

Final reports often contain correct claims with missing or wrong citations. Traces show the web search and document analysis subagents return prose summaries, and source URLs and page numbers are mixed into the text inconsistently or dropped when findings are condensed. Which two changes best preserve attribution through to the final report? Select two.

Show answer and explanation

Answer: B and C — Require subagents to output structured claim-source mappings (claim, source URL or document name, page, relevant excerpt) that the synthesis subagent must preserve and merge. / Pass context between agents in a structured format that separates content from metadata such as source URLs, document names, and page numbers, rather than embedding them in prose.

Attribution gets lost when findings are compressed without claim-source mappings. Requiring structured claim-source mappings that synthesis must preserve (b) and passing context in formats that separate content from metadata (c) keep provenance intact end to end (tasks 5.6 and 1.3). Searching for citations afterwards (a) attaches sources that may not support the claim, which is worse than having none. A larger context window (d) doesn't give the content any structure, and verbose originals make the downstream context budget harder to manage.